Meet the Technical Product Manager who is helping shape the next generation of CoCounsel Legal into a more robust product for lawyers
Highlights
- As a Technical Product Manager, Nina Koteva helps guide and shape the next generation of CoCounsel Legal.
- Continuous testing and customer feedback shaped the next-gen legal AI assistant.
- Rebuilding from scratch balanced trustworthiness with speed for legal workflows.
Nina Koteva started her journey at Thomson Reuters as an Applied Research Scientist at Thomson Reuters Labs. Five years later, she moved into a Technical Product Manager role, where her work centered on internal systems. Today, she helps guide the build of what may be the most ambitious product in the company’s legal AI portfolio: the next generation of CoCounsel Legal.
Her path on this product mirrors the product’s own arc. What started as a proof of concept moved into alpha, then beta, with each phase raising the stakes.
“Originally, [CoCounsel] was a proof of concept, so we were working on a 0 to 1 version of the next generation legal assistant,” Koteva says. “Afterwards, once we proved out that concept, it turned into getting it into the hands of a small handful of customers. From there, it turned into going into beta and actually turning it into a product.”
Today, her focus spans CoCounsel Legal’s core agent experience and the evaluation systems behind it, making sure the product performs reliably across the full range of tasks lawyers bring to it every day.
Jump to ↓
Building for how lawyers actually work
The lessons learned from rebuilding
Moving from concept to reliability
Building for how lawyers actually work
Scale was one of the first real tests Koteva encountered. Early in development, the CoCounsel product team built more than 500 use cases to put the system through its paces. It was a strong foundation, but it fell short of what the work required.
“We saw that even that is not enough for the variety and diversity of tasks that lawyers actually need to do,” Koteva says.
The team responded by expanding scope significantly, working with subject matter experts across Thomson Reuters to add depth in areas that matter to real legal practices like employment law, internal investigations, and government-related work.
That coverage work directly shapes what CoCounsel can do. The broader the evaluation set, the more confidently the agent can handle real-world requests across practice areas.
Feedback as a design tool
Rather than waiting for a later stage to learn what was working, the team built continuous testing into CoCounsel’s development from the start.
“What we realized very early in the process is that having a platform to do live testing got us feedback the fastest and helped us make progress very rapidly.”
Technical Product Manager, Thomson Reuters
Early prototypes gave internal experts and select customers a chance to interact with the system in real time. As the next-gen build moved into beta, that feedback loop widened considerably. Customers were using it in real scenarios and actively influencing where it went next.
“We got tons of valuable feedback from them and tried to incorporate it as quickly as we could back into the core agent or the application experience,” she says.
That input is actively shaping what comes next. Tester insights are driving meaningful enhancements across the experience, from making interactions feel more natural and intent-driven to bringing key workflows like document review and citation exploration fully within the product. These cases reflect how lawyers work, and they are shaping the roadmap in real time.
The lessons learned from rebuilding
One of the most consequential decisions was to rebuild CoCounsel Legal from the ground up. That choice created room for meaningful improvements, along with the team carefully thinking about speed. Legal work demands outputs that can be trusted, not just outputs that arrive fast. A single request — research, then drafting, then document comparison — can require several sequential steps. When the system also verifies citations before surfacing results, that adds time.
“How we balance that type of trustworthiness in the outputs with latency has been another area [for improvement],” Koteva says. There is no shortcut here. Getting it right matters more than getting it done quickly.
Moving from concept to reliability
For Koteva, watching CoCounsel Legal move from an early idea into something lawyers use in their daily work has been the most rewarding part.
“I’ve been with this product since its very inception, so seeing it go into the hands of customers and become a real product with end-to-end journeys has been incredibly exciting,” she says.
In fact, the testing feedback was so compelling that Thomson Reuters accelerated CoCounsel Legal’s move to early access ahead of schedule. This is a testament to how confidently the product delivered against expectations.
As CoCounsel Legal shifts into early access, the next phase brings broader usage and, with it, a much wider view of how the product performs across real firms and practice areas.
“I think the other thing [to be excited for] is to see the feedback of our customers at a much bigger scale,” Koteva says.
That broader signal — from more firms, more practice areas, and more types of legal work — is where the real test begins. For a product that has come this far this fast, it is exactly the kind of challenge the team has been building toward.
The next generation of CoCounsel Legal is scheduled to launch later this year. Learn more about how we’re building the future of legal AI.
